Mapping the HLA-DO/HLA-DM complex by FRETand mutagenesisTaejin Yoona,1, Henriette Macmillana, Sarah E. Mortimerb, Wei Jianga, Cornelia H. Rinderknechta,2, Lawrence J. Sternb,and Elizabeth D. Mellinsa,3

aDepartment of Pediatrics, Program in Immunology, Stanford University, Stanford, CA 94305; and bDepartment of Pathology, University of MassachusettsMedical School, Worcester, MA 01655

Edited by Daved H. Fremont, Washington University School of Medicine, St. Louis, MO, and accepted by the Editorial Board June 1, 2012 (received for reviewAugust 25, 2011)

HLA-DO (DO) is a nonclassic class II heterodimer that inhibits theaction of the class II peptide exchange catalyst, HLA-DM (DM), andinfluences DM localization within late endosomes and exosomes.In addition, DM acts as a chaperone for DO and is required for itsegress from the endoplasmic reticulum (ER). These reciprocalfunctions are based on direct DO/DM binding, but the topologyof DO/DM complexes is not known, in part, because of technicallimitations stemming from DO instability. We generated twovariants of recombinant soluble DO with increased stability[zippered DOαP11A (szDOv) and chimeric sDO-Fc] and confirmedtheir conformational integrity and ability to inhibit DM. Notably,we found that our constructs, as well as wild-type sDO, are in-hibitory in the full pH range where DM is active (4.7 to ∼6.0). Toprobe the nature of DO/DM complexes, we used intermolecularfluorescence resonance energy transfer (FRET) and mutagenesisand identified a lateral surface spanning the α1 and α2 domainsof szDO as the apparent binding site for sDM. We also analyzedseveral sDMmutants for binding to szDOv and susceptibility to DOinhibition. Results of these assays identified a region of DM im-portant for interaction with DO. Collectively, our data define a pu-tative binding surface and an overall orientation of the szDOv/sDM complex and have implications for the mechanism of DO in-hibition of DM.

antigen processing | antigen presentation | MHC class II

Major histocompatibility complex (MHC) class II moleculespresent peptides on the antigen-presenting cell (APC)

surface to sensitize CD4+ lymphocytes. The class II presentationpathway is well-characterized and includes roles for three ac-cessory molecules: invariant chain (Ii), HLA-DM (DM), andHLA-DO (DO) (1). Newly synthesized MHC class II moleculesbind to Ii in the ER. A trafficking signal in the cytoplasmic do-main of invariant chain directs the complex to late endosomalcompartments (2, 3), called MIIC (MHC II containing com-partments). The invariant chain (Ii) is proteolytically degraded inthis low-pH environment, leaving a nested set of invariant chainfragments, CLIPs (class II-associated invariant chain peptides),in the class II peptide-binding groove (4). CLIP is ultimatelyexchanged for antigenic peptides by DM, a nonclassic MHC classII molecule, which further influences peptide selection in a pro-cess termed “peptide editing” (5–10). DM also stabilizes emptyMHC class II to maintain a peptide receptive structure; other-wise, empty MHC class II proteins become peptide averse (11).DO, another nonclassic MHC class II αβ heterodimer, first

was described as an inhibitor of DM function, because over-expression of DO in human cells increased levels of CLIP-boundclass II molecules (12, 13). In vitro evidence most often cor-roborated the view of DO as a negative modulator of DM (12,14). Results suggesting that DO inhibition of DM is robust atearly endosomal pH (6.0–6.5) and attenuated at the lower pH(4.5–5.0) of late endosomal/lysosomal MHC II containing com-partments (MIICs) (14) raised the possibility that DO modifiesDM-mediated peptide exchange by limiting the location of fully

functional DM. This idea was consistent with evidence that DOdecreased presentation of antigens taken up by fluid-phase en-docytosis and enhanced BCR-mediated antigen presentation (14).DO does not unequivocally facilitate antigen presentation by allBCRs, however (15).DO has been reported to have several functions. It mediates

redistribution of DM from internal vesicles to the limiting mem-brane of multivesicular bodies (16) and contributes to reducedDM sorting to exosomes (17). Constitutive overexpression of DOin dendritic cells blocks disease in the nonobese diabetic (NOD)mouse model of type 1 diabetes (18). Finally, DO expression inmurine B cells limits their participation in the germinal centerreaction through effects on antigen presentation (19). However,the full physiological function of DO is still unclear.In human B lymphocytes, the majority of DO is in complex

with DM. DO/DM complexes are capable of egress from the ER(12, 20–22), whereas DO alone is not (23), most likely because ofimproper folding (24). DO instability has made functional,structural, and biochemical studies difficult. However, Thibo-deau and coworkers found a point mutation (DOαP11V) thatallows DO to independently egress the ER (23); this mutation isin a sequence stretch predicted by homology to influence α/βpairing. Mutation of DOαP11 allows expression in insect cells ofmore stable, secreted DO molecules for biochemical studies. Wecharacterized the behavior of these molecules in DM inhibitionassays, and we used in vitro fluorescence resonance energytransfer (FRET) assays and mutagenesis to study DO/DMbinding orientation. Our findings have implications for the roleand mechanism of action of DO.

ResultsProduction and Characterization of Recombinant Soluble DO. UsingS2 insect cells, we expressed the luminal domains of DOα andDOβ, with additions of an αP11A mutation and C-terminalcomplimentary leucine zippers for increased stability, and epi-tope tags for purification (Fig. S1 A and B). The final yield was0.5–1.0 mg/L insect cell supernatant, similar to MHC class IIproteins such as DQ or I-A (Fig. S1C). Purified recombinant

Author contributions: T.Y., H.M., S.E.M., L.J.S., and E.D.M. designed research; T.Y., H.M.,S.E.M., and W.J. performed research; C.H.R. contributed new reagents/analytic tools; T.Y.,H.M., S.E.M., and W.J. analyzed data; and T.Y., H.M., S.E.M., L.J.S., and E.D.M. wrotethe paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission. D.H.F. is a guest editor invited by theEditorial Board.

Freely available online through the PNAS open access option.1Present address: Department of Chemistry, Yonsei University, Seoul 120-749, Republicof Korea.

2Present address: Department of Pathology, Genentech Inc., South San Francisco,CA 94080.

3To whom correspondence should be addressed. E-mail: mellins@stanford.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1113966109/-/DCSupplemental.

11276–11281 | PNAS | July 10, 2012 | vol. 109 | no. 28 www.pnas.org/cgi/doi/10.1073/pnas.1113966109

soluble, modified DO (hereafter “szDOv”) contained a singleglycan at position DOα79N (Fig. S1 D–F). To test whetherszDOv protein was functional, we measured its capacity to in-hibit sDM in endpoint class II peptide-loading assays. szDOvinhibited peptide exchange by sDM in a concentration-de-pendent manner, whereas two control proteins, IgG Fab (frag-ment antigen-binding region) and BSA, did not (Fig. 1A). Thefunctionality of szDOv molecules and their conformational in-tegrity (confirmed by precipitation with a conformation sensitivemAb; see Fig. 3B) supported their use in further studies. We alsoused S2 insect cells to prepare sDO-Fc, with the DOα and -β Ctermini fused to human IgG1 Fc (crystallizable fragment)domains for stabilization, as described previously (14). DO wasalso expressed with DM in insect cells, and preformed DO/DMcomplex was purified from supernatants. The DO/DM complexwas very stable, with slow dissociation over months at 4 °C (Fig.S1G). Following dissociation from DO, DM regained full activityin peptide association assays (Fig. S1H).

Soluble DO: pH-Independent Inhibition. It has been reported thatthe inhibition of DM by sDO-Fc is strikingly pH-dependent, withlittle or no inhibition at pH < 5.5 (14). Differently from this, wefound that szDOv inhibits DM function across a pH range of4.7–6.0 (Fig. 1B). At pH 7.0, DM peptide-loading activity was toolow to evaluate inhibition, as expected based on prior datashowing pH 4.5–6.0 as an optimal pH range for DM catalyticfunction (14, 21). The lack of pH dependence of DO activitypotentially could have been attributable to the introduction of

the αP11A mutation or the leucine zipper (Fig. S1A). To testthis, we used sDO/sDM in complex or sDO-Fc (both withoutthe αP11A mutation) and measured DM inhibition at pH 4.5–7.0. We used fluorescence polarization (FP) for real-timemeasurement of peptide/DR association. DO inhibition ofsDM was effective at pH 4.5–6.0 using sDO-Fc (Fig. 1C). Toverify that DO was acting on DM, and not inhibiting DR/peptide association directly, we allowed the reactions in Fig. 1Cto come to equilibrium, because the catalytic action of DMinfluences the kinetics of DR/peptide interaction but not theequilibrium. At 5 d, all reactions had levels of DR1/peptidesimilar to that of sDR1 and peptide alone (Fig. 1D). Thus, DOmolecules reduce the levels of complex in the shorter timeframe by affecting DM catalytic action. We also studied pre-formed sDO/sDM complex over the same pH range (Fig. 1C).No increase in DM activity was observed for sDO/sDM atacidic pHs; DO inhibited DM completely whether DO wasadded to sDM at the beginning of the experiment or as pre-formed sDO/sDM. Additionally, pH does not affect the sta-bility of preformed DO/DM complex, because complexintegrity was unchanged following 4 d at pH 5 and 7 (Fig. S1F).Taken together, our data indicate that sDO (with or withoutthe αP11A mutation and with or without Fc and leucine zipper)inhibits DM function in a pH-independent manner.

Side-by-Side Orientation of DO/DM Complex Revealed by FRET. It hasbeen proposed that DOαE41 and a region C terminal to DOα80are key to DO/DM interaction (23); however, the overall struc-ture of DO/DM complex is not well defined. We used in vitrointermolecular FRET to elucidate the binding orientation ofDM and DO. FRET relies on the transfer of energy from a do-nor molecule (dye or fluorophore) to an acceptor molecule (dyeor fluorophore); the absorption spectrum of the acceptor mustoverlap the fluorescence emission spectrum of the donor. Theenergy transfer reduces the fluorescence intensity and excitedstate lifetime of the donor molecule and increases the emissionintensity of the acceptor molecule. The efficiency of the processdepends on the inverse sixth-distance between donor and ac-ceptor. By measuring FRET between donor fluorophores in-troduced at different sites on DO and an acceptor fluorophoreon DM, we were able to obtain information on the relative ori-entation of DO and DM in the complex.To choose sites to introduce cysteines for dye labeling, we first

modeled the DO structure, using the sDR1 structure (25) asa template in the alignment mode of SWISS-MODEL (Fig. 2A).We identified solvent-exposed residues to minimize the effects ofcysteine introduction and dye labeling on the tertiary structure ofszDOv. We generated four szDOv mutants, each with a freecysteine (αA62C, αR80C, αE131C, and βS63C), and labeledthem with maleimide-Cy3 as fluorescence donors (Fig. 2A, Left).The mutations are distributed over szDOv, allowing assessmentof multiple regions of the molecule. Recombinant sDM hasa single free cysteine at DMβ46, the mutation of which does notreduce DM function in peptide loading (26). This residue islikely to be close to both DR- and DO-binding surfaces, becausean aberrant N-glycan at DMβN108, near sDMβC46 (Fig. 2A,Right), interferes with DM/DR and DO/DM interactions (26).Therefore, we labeled DMβC46 with maleimide-Cy5, a fluores-cence acceptor. The calculated dye/protein ratios for sDM andszDOv mutants (based on the UV visible spectrum) were in therange of 0.90–0.95 dye/protein. To determine whether dye la-beling disrupted DO function, we measured DM inhibition byCy3-labeled szDOv mutants in peptide-loading assays. The la-beled DO mutants exhibit concentration-dependent inhibition ofDM comparable with that observed with szDOv (Fig. S2A). Wealso examined the apparent dissociation constants of unlabeledszDOv and Cy3-labeled (αE131C) szDOv with sDM using theOctet biosensor system (details in SI Materials and Methods). The

A B[sDM] = 0 nM, [protein] = 0 nM[sDM] = 400 nM, [protein] = 0 nM[sDM] = 400 nM, [protein] = 100 nM[sDM] = 400 nM, [protein] = 200 nM[sDM] = 400 nM, [protein] = 400 nM

[sDM] = 50 nM, [szDOv] = 0 nM[sDM] = 50 nM, [szDOv] = 100 nM[sDM] = 50 nM, [szDOv] = 200 nM[sDM] = 50 nM, [szDOv] = 400 nM

Fold

cha

nge

over

400

nM s

DM

alo

ne

Fold

cha

nge

over

50 n

M s

DM

at p

H 51.2

10.80.60.40.2 0

1.2 10.80.60.40.2 0

10.80.60.40.2 0

Fold

cha

nge

over

sDR

1+ s

DM

pH

5

pH 4.5 5.0 5.5 6.0 6.5 7.0 pH 4.5 5.0 5.5 6.0 6.5 7.0

mP

250200150100 50 0

C DsDR1 sDR1+sDMsDR1+sDO/sDM sDR1+ sDM + sDO-Fc

szDOv Fab BSA pH 4.7 5.0 6.0 7.0

No sDR1 sDR1 sDR1+sDMsDR1+sDO/sDM sDR1+ sDM + sDO-Fc

Fig. 1. pH-independent inhibition by recombinant soluble DO. (A) In-hibition of DM-mediated peptide loading by szDOv at pH 5. Binding of HA-biotin peptide to sDR4-CLIP with or without sDM and the indicated amountsof szDOv or control proteins was measured by capture ELISA. Data, from oneof two experiments with similar results, are expressed as normalized foldchange, with enhancement of peptide loading with sDM alone normalizedto 1. (B) Inhibition of DM-mediated peptide loading by szDOv at pH 4.7–7.0.Capture ELISA data, from one of two experiments with similar results,showing loading of HA-biotin peptide on sDR4-CLIP with sDM in the pres-ence of titrated amounts of szDOv at the indicated pH and expressed asnormalized fold change, with loading by sDM alone at pH 5 normalized to 1.(C) Inhibition of DM-mediated peptide loading by sDO-Fc at pH 4.5–7.0.sDR1 was incubated with or without sDM, sDO-Fc, or sDO/sDM complex incitrate (pH 4.5–6), 2-(N-morpholino)ethanesulfonic acid (Mes) (pH 6.5), orHepes (pH 7). Reactions were initiated by adding Alexa-488-HA peptide, andFP was measured over time. Changes in initial rate (determined from linearfits to early portions of FP change against time) were measured. Data,representative of three experiments, are fold change in initial rate ofloading ± SE, relative to loading with sDM at pH 5.0. (D) sDO inhibition ofsDM-mediated peptide loading at pH 4.5–7.0 after 5 d. FP values (mP) after5 d of incubation of indicated proteins at indicated pH are shown.

Yoon et al. PNAS | July 10, 2012 | vol. 109 | no. 28 | 11277

IMMUNOLO

GY

binding between Cy3-labeled DO and sDM (KD ≤ 5.00 nM) wasof comparable strength to that between unlabeled szDOv andsDM (KD∼3.72 nM) (Fig. 2B). A total of 99% of donor dye-la-beled DO molecules (100-300 nM) would theoretically be en-gaged in DO/DM complexes under conditions of 5× excess ofacceptor dye-labeled DM molecules (500–1,500 nM). Takentogether, these results showed that the Cy3-labeled DO mutantswere suitable for in vitro FRET experiments.Fluorescence emission spectra were acquired at 25 °C, with

excitation at 500 nm and a 550-nm emission long-pass filter.Donor only samples (Cy3-szDOv) and an acceptor only sample(Cy5-sDM) present distinct emission maxima at 570 and 670 nm,respectively. We used a donor quenching approach, with transferof energy from donor to acceptor reducing intensity in the donorfluorescence signal monitored at the emission maximum. Thus,donor emission decreases with increasing FRET, with the FRETefficiency differing, depending on the distance between donor(Cy3) and acceptor (Cy5).Shown in Fig. 2C are the fluorescence emission spectra of DO/

DM complexes with DO labeled at αR80C, αE131C, and βS63C.FRET efficiency appeared greatest when the donor was locatedat αE131C (at 10× Cy5-sDM: ∼0.44) and least when the donorwas located at αR80C (at 10× Cy5-sDM: ∼0.17). In the fluo-rescence emission spectra of αA62C, unlike the others, donorfluorescence intensities are not proportionally decreased withincreased acceptor fluorescence intensity. With the αA62C do-nor, acceptor fluorescence intensity at 2× Cy5-sDM (Fig. 2C,green) looks comparable with that of the αE131C donor at 10×Cy5-sDM (Fig. 2C, gray). However, donor intensity of αA62C isreduced by <50% of the reduction in intensity of αE131C, sug-gesting that αA62C gives anomalous FRET (see Discussion),and, therefore, αA62C results are excluded from distance anal-ysis. Nonspecific FRET (either MHC class I-Cy5/sDM-Cy3 orMHC class I-Cy5/szDOv-Cy3) occurred at <10% of DO/DMFRET under the same conditions (Fig. S2B), arguing that DO/DM FRET results derive from the specific interaction betweenDO and DM.In Fig. 2D, FRET efficiencies of the three informative

donors are plotted as a function of fold excess of acceptor

(Cy5-sDM). FRET efficiencies increased with increasing ac-ceptor concentrations and started to saturate at 5× excessconcentration of acceptor, as expected from dissociation con-stants. FRET efficiencies were measured in the presence of upto 10× excess acceptor, and distances (R) are derived from:R = R0 (FRET−1 − 1)1/6, where R0 = 53 Å (for Cy3/Cy5).Collectively, the FRET results suggest that the lateral surfacecontaining szDOvαE131C is in close proximity to sDM in theDO/DM complex.

Aberrant Glycan at DOα128 Interrupts DO/DM Interaction. To testthe region around αE131C for interaction with DM, we in-troduced an aberrant N-glycan at szDOvα128 (Fig. 3A, mutantT128N). Glycan addition was confirmed by reduced α-chainmobility on SDS/PAGE (Fig. S1I and Fig. 3B). The intact overallstructural integrity of the szDOvαT128N mutant was indicatedby its recognition by a conformation-sensitive anti-DO mAb,Mags.DO5 (24), in immunoprecipitation assays (Fig. 3B). In-hibition of sDM by the szDOvαT128N mutant is significantlyreduced in peptide-loading assays (Fig. 3C). Consistent with thisresult, the apparent KD between sDM and szDOvαT128N is>20× higher than that of szDOv, based on equilibrium bindinganalysis (Fig. 3D). Thus, both FRET data and mutational anal-yses argue that the surface of szDOv containing residues αT128and αE131 is oriented toward DM in the szDOv/sDM complex.We also examined the effect of removing the N-glycan at

DOαN79, using mutant szDOvαS81A (Fig. S1I). Immunopre-cipitation of szDOvαS81A protein by Mags.DO5 (Fig. 3B)indicated its overall structural integrity. However, size-exclu-sion chromatography showed the protein was highly aggre-gated (Fig. S1J), suggesting an essential role of the glycan inpreventing aggregation.

Electrostatic Interaction in the DO/DM Complex. In the DO struc-tural model, the lateral surface we identified as the putative DM-binding region on DO includes an exposed loop with severalcharged residues in the α1 domain. DOαE41, located on thisloop, was previously reported as important for DM binding, ERegress, and DO activity (23). In addition, the side chains of two

DO DM

βC46

βN108-CHO(βR110S)

βS63

αA62

αR80

αE131

A

C D

B

[szDOv] nM [αE131C-Cy3] nM

KD=3.72±0.69 nM

R2= 0.9857

KD=4.45±1.21 nM

R2= 0.9699

Rel

ativ

e in

tens

ity (a

rb. u

nit) 1.0

0.5

01.0

0.5

0

FRE

T ef

ficie

ncy

Molar excess of sDM-Cy5

4.0

2.0

0nm S

hift

(nm

)

Time (sec)0 10000 20000 0 500 1000

4.0

2.0

0

Max

nm

Shi

ft (n

m)

0 500 1000Time (sec)

0 10000 20000

4.0

2.0

0nm S

hift

(nm

) 4.0

2.0

0

Max

nm

Shi

ft (n

m)szDOv αE131C-Cy3

0 fold sDM-Cy5 0 fold sDM-Cy52 fold sDM-Cy5 10 fold sDM-Cy5

10 fold sDM-Cy510 fold sDM-Cy5

0 fold sDM-Cy5 0 fold sDM-Cy5

αE131C (10- fold sDM-Cy5)

αA62C αR80C

βS63C αE131C570 620 670 570 620 670 nm

0 5 10 0 5 10

0.6

0.4

0.2

0.00.6

0.4

0.2

0.0

FRE

T ef

ficie

ncyαA62C αR80C

βS63C αE131C

AnomalousFRET

FRET=0.17(69Å)

FRET=0.26(63Å)

FRET=0.44(55Å)

Fig. 2. In vitro FRET with szDOv and sDM. (A) Locations of introduced free Cys on DO (blue circles) and natural free Cys on DM (red circle). The DO structurewas simulated, based on HLA-DR1 (Protein Data Bank ID code 1AQD): DOα (cyan); DOβ (purple). The DM structure was visualized using WebLab Viewer LiteV3.7: DMα (cyan); DMβ (purple). Glycan addition site (DMβN108) in mutant DMβR110S is shown. (B) Octet analysis of unlabeled szDOv and szDOvαE131C-Cy3binding to immobilized sDM-biotin. Concentrations of szDOv and szDOvαE131C-Cy: 0.24 (blue), 0.98 (brown), 3.9 (pink), 15.6 (cyan), 62.5 (sky blue), 250 (red),and 1,000 nM (black). Plots of maximum wavelength (nm) shift under equilibrium binding conditions against increasing concentrations of DO were used todetermine KD. (C) FRET spectra of Cy3-labeled szDOv mutants and Cy5-labeled sDM. Cy3-labeled szDOvαA62C, szDOvβS63C, αR80C, and αE131C were in-cubated with up to 2× excess (szDOvαA62C) or up to 10× excess (others) Cy5-labeled sDM (red), as described in SI Materials and Methods. Spectra wereobtained with buffer alone (without szDOv-Cy3), titrated with the same concentrations of sDM-Cy5, and subtracted from the corresponding FRET spectrum.(D) Distances between the Cy3 donor attached at different positions on szDOv (szDOvαA62C-Cy3 excluded) and the Cy5 acceptor attached at sDMβC46 werecalculated: R = R0 (FRET−1 − 1)1/6, where R0 represents 53 Å (for Cy3/Cy5). Data from one of two experiments with similar results are shown.

11278 | www.pnas.org/cgi/doi/10.1073/pnas.1113966109 Yoon et al.

lysines (DOα K38,K39) near the DOαE41 residue are predicted(in the DO model) to be solvent accessible and in the plane ofthe putative DM-binding face of DO (Fig. 3A). To evaluate theeffect of these charged residues on szDOv/sDM interaction, wemade a szDOvαE41K mutant, as well as two double mutants:szDOvαK38,39A and szDOvαK38,39D. These mutant moleculeswere conformationally intact, based on recognition by Mags.DO5 antibody (Fig. 3E). In comparison with szDOv, two of threemutants were less capable of inhibiting sDM in peptide-loadingassays (Fig. 3F). szDOvαK38,39D is more defective thanszDOvαK38,39A, with comparable function with szDOvαE41Kin this assay. These findings argue that a lysine from αK38,K39and the glutamic acid αE41 are among the residues involved inelectrostatic interactions between szDOv and sDM and furtherimplicate this region in szDOv binding to sDM.

Effect of Mutations in the DMβ2 Domain on Interaction with DO.Previously, we showed that full-length DMβR110S, which hasan aberrant N-glycan at DMβN108, substantially reduces bothDM/DR and DO/DM complex formation in human B cells (26).We, therefore, tested formation of szDOv/sDM complexes withsoluble DM carrying this mutation (sDMβR110S). The ability ofszDOv to inhibit peptide exchange by sDMβR110S could not bemeasured, as functional interaction with class II is severelycompromised by this DM mutation (Fig. S2C). We measured

direct binding between sDMβR110S and szDOv, using theOctet. The binding of wild-type sDM (sDMwt) to immobilizedszDOv exhibits fast association and slow dissociation, whereassDMβR110S associates more slowly and dissociates quickly(Fig. 4A), showing that the added glycan interferes directly orindirectly with szDOv/sDM interaction. Equilibrium bindinganalysis showed a reduction in the apparent KD of interactionbetween sDMβR110S and szDOv compared with sDMwt(Fig. 4B).We interrogated other residues in the DMβ2 domain by

making additional mutations. We found that the DO inhibitioncapacity of sDMβH141A,S142A is significantly reduced (Fig.4C), and the apparent KD of binding to DO is modestly reduced(Fig. 4B). We also tested mutant sDMβE177N,I179T, which hasan aberrant N-glycan at DMβ177 (Fig. S2D). This mutation didnot affect the capacity of the mutant protein to be inhibited byDO (Fig. 4C) or to bind to DO (Fig. 4B), implying it is locatedoutside the DO-binding site on DM. Our data suggest that theszDOv-binding surface on sDM includes the area aroundDMβN108 (the N-glycan location in sDMβR110S) and is influ-enced by the double mutation DMβH141,S142 but does notextend to sDMβE177,I179. We tested several other DM mutantslocated on the putative DO-binding surface and, consistent withFig. 4 data, those show reduced interaction with DO, whereasmutants located farther from the putative DM/DO interactionsurface interact well with DO (Table S1). Based on all our data,we propose an orientation of szDOv/sDM complex with theoverall topology shown in Fig. 4D.

DiscussionDO, a nonclassic MHC class II molecule, is involved in the an-tigen presentation pathway as a modulator of DM (12, 13). Al-though data show that DO can inhibit peptide exchangecatalyzed by DM, the physiological role of DO remains un-known. More detailed structural and biochemical information islikely to lead to better understanding of DO function. However,study of DO has been hindered by the instability of soluble formsof this protein. Here, we introduced a mutation at DOαP11(P→A) that closely resembles a variant (P→V) first described byThibodeau and coworkers (23). Removal of the proline at theDOα11 position facilitates production of more stable, recombi-nant, soluble DO molecules that inhibit DM in class II peptide-loading assays.Our FRET results constrain possible orientations of DO/DM

complex, and the one most consistent with the FRET datapositions the lateral surface containing DOαE131 toward DM.This putative DM-binding surface on DO also is consistent withour mutagenesis results (introduction of an aberrant N-glycan atDOα128 and mutation of several charged residues in the α1domain). Corroborating our findings, Thibodeau and coworkersreported that chimeric DO/DR molecules implicate DOα in in-teraction with DM, and they also identified DOα41 as a criticalresidue (23). Our data from functional assays and in vitro bindingmeasurements of DMβ mutants suggest that a lateral face of DM(between DMβ108 and DMβ141,142) contributes to or influen-ces the binding surface for DO.Our overall DO/DM orientation model provides possible

explanations for the anomalous FRET at αA62C. According toour model, αA62C-Cy3 is located near the interface betweenDO/DM. This could lead to anomalous Cy3 fluorescence en-hancement or could restrict motion of the Cy3 molecule.Anomalous Cy3 fluorescence occurs when the environmentsurrounding the protein bound Cy3 dye becomes hydrophobic,because the fluorescence quantum yield (Φ) of Cy3 is dramati-cally increased with increasing viscosity (27, 28). Thus, thediminished reduction in donor fluorescence intensity of αA62C-Cy3 compared with αE131C-Cy3, despite the comparable ac-ceptor intensities of the two Cy3 probes at 2× and 10× sDM-Cy5,

AE41

K38,39

T128

anti-

FLAG

M

ags.

DO

5 an

ti-FL

AG

Mag

s.D

O5

anti-

FLAG

M

ags.

DO

5

anti-

FLAG

M

ags.

DO

5

S2

szD

Ov

szD

Ov

szD

Ov

szDOv

0

0.2

0.4

0.6

0.8

1

1.2

!"#$%&''()* !"#$%&''()+,'-./*

Fol

d ch

ange

ove

r 4

00 n

M D

M, n

o sz

DO

v

[DO]******

szDOv szDOv T128N

0

1

!"#$%&''()* +,-'.* +./01/2(* +./01/2#*

[DO]

szDOv K38,39A K38,39D

szDOv

* *

anti-

FLAG

M

ags.

DO5

anti-

FLAG

M

ags.

DO5

anti-

FLAG

M

ags.

DO5

anti-

FLAG

M

ags.

DO5

S2

szD

Ov

E41K

szD

Ov

K38,

39A

szD

Ov

K38,

39D

D nM2

D

EszDOv T128N

1.2

0.6

0

1anti-FLAG2Mags.DO512121212

S2

szD

Ov

T128

N

S81

A

szDOv

0 1250 2500[szDOv T128N] nM

Time (sec)

3.0

1.5

0

3.0

1.5

0

nm S

hift

(nm

)M

ax n

m S

hift

(nm

)

KD=82.15±11.94 nM

R2=0.9873

Fold

cha

nge

over

400

nM

sD

M, n

o sz

DO

v

szDOv E41K K38,39A

12121212

S2 E

41K

K38

,39A

K38

,39D

szDOv

K38,39D

1anti-FLAG2Mags.DO5

F

Fold

cha

nge

over

400

nM

sD

M, n

o sz

DO

v

B

C

Fig. 3. Aberrant N-glycan at DOα128 interferes with szDOv/sDM in-teraction; charged residues are involved in szDOv/sDM interaction. (A)Model wt DO structure with DOα (cyan) and DOβ (purple) with siteschosen for mutation: szDOvαΤ128, positively charged szDOvαK38,39 (red),and negatively charged szDOvαE41 (green). (B) Structural integrity ofszDOvαT128N and szDOvαS81A mutants. szDOv, szDOvαT128N, andszDOvαS81A were immunoprecipitated with indicated mAb and detectedwith anti-DOα rabbit polyclonal antisera. S2: untransfected S2 cells wereused as a negative control. (C) Inhibition of DM-mediated peptide loadingby szDOvαT128N. Peptide-loading data (from one of three experiments withsimilar results), collected as in Fig. 1A. Statistics were obtained by two-wayANOVA, compared with szDOv (Graphpad Prism5). ***P < 0.001. (D) Octetanalysis of szDOvαT128N binding to immobilized sDM-biotin. (Upper) Bind-ing of szDOvαT128N [at 5 nM (blue), 50 nM (brown), 100 nM (pink), 150 nM(cyan), 300 nM (sky blue), 1,000 nM (red), and 2,500 nM (black)]. (Lower)Plot of maximum wavelength (nm) shift under equilibrium binding con-ditions against increasing concentrations of szDOvαT128N. (E) Structuralintegrity of szDOvαE41K and szDOvαK38,39A/D mutants. szDOvαE41K andszDOvαK38,39A/D immunoprecipitated with indicated mAb and detectedwith anti-DOα rabbit antisera. S2: untransfected S2 cells were used asa negative control. (F) Inhibition of DM-mediated peptide loading by szDOvmutants. szDOvαE41K, K38,39A, and K38,39D were incubated with sDM,sDR, and peptide, as in Fig. 1A. Data are from one of two experiments withsimilar results. Statistics were obtained by two-way ANOVA, compared withszDOv. *P < 0.05.

Yoon et al. PNAS | July 10, 2012 | vol. 109 | no. 28 | 11279

IMMUNOLO

GY

respectively, could reflect an increase in fluorescence quantumyield of αA62C-Cy3 in the presence of sDM. Alternatively, re-stricted molecular motion of the Cy3 at αA62C in the presenceof sDM could induce the failure of the ideal dipole approxima-tion, which is implicit in the Förster Eq. (29). In either case, DMitself or a conformational change in DO induced by DM influ-ences the environment of DOαA62. Using Mags.DO5 antibody,Thibodeau and coworkers have shown that full-length DMinfluences the conformation of full-length DO in cells (24). TheMags.DO5 epitope is located on the DOβ chain, whereas our datasuggest DM-mediated effects on DOα. The DM-mediated con-formational changes in DO, including the effect on the environ-ment of αA62 in szDOv, likely contribute to the DM chaperoningknown to be required for efficient ER egress of wt DO (23).Data from Karlsson and coworkers showed pH-dependence in

DO inhibition of DM: DO efficiently blocked the peptide-editingfunction of DM at mild acidic pH (5.5–6.5) but not at lower pH(4.5–5.0) (14). This in vitro observation, along with severalresults from the H2-DO-deficient mouse (14), supported a hy-pothesis in which DO limits the pH range of DM function and,thereby, favors presentation of B-cell receptor (BCR)-boundantigen compared with antigen taken up by fluid-phase endocy-tosis. In contrast to previous results using recombinant DO, wefind that several forms of recombinant, soluble DO molecules(szDOv, sDO-Fc, and sDO) inhibit peptide exchange by DMthroughout the range of pH 4.5 to ∼6.0. As a consequence, DOwould be expected to inhibit DM function deep into the endo-cytic pathway. Consistent with our findings, Denzin et al. alsoreported pH-independent DO inhibition with full-length DO/DM (30). A second group also observed DO inhibition of DM atpH 4.5–6.0, using DO/DM complexes from MelJuso cells (21).However, they used the term “pH-dependent” to emphasizerelatively strong inhibition at pH 6 after 24 h of incubation. It ispossible that the DO/DM complexes in their experiment becameunstable at pH 4.5 to ∼5.0 after 24 h of incubation and thatpartial dissociation of DO improved DM activity under theseconditions. This possibility would be consistent with our ownobservation that, upon prolonged storage of sDO/sDM com-plexes, we observed recovery of some DM activity, presumablyfrom free DM (Fig. S1H).MHC class II molecules and DO share conserved N-glycan

motifs that differ from DM (Fig. S3). The N-glycans on class IIare thought to play a role in proper folding of nascent molecules(31), in protease resistance (32), in facilitating optimal receptorgeometry for APC/T-cell communication (33), and, most re-cently, in presentation of particular carbohydrate antigens (34).We found that the α1 domain N-glycan at szDOvαN79, nearα-chain/β-chain interface and homologous to αN78 in conven-tional MHC class II, influences the aggregation state of themolecule (Fig. S1J). The homologous region of DM, whichnaturally lacks this N-glycan, has been implicated in interactionwith MHC class II by mutagenesis (26). Compared with con-ventional class II and DO, DM has a unique N-glycan atDMβN92, located on the opposite side of the protein, near its α/βinterface (35). This DM surface also might be prone to protein/protein interaction were it not inhibited by the presence of theglycan. Indeed, our data suggest that the homologous surfaces onMHC class II (36) and szDOv, which lack this N-glycan, are usedfor DM binding.In our predicted overall orientation of szDOv/sDM complex,

the binding surface for szDOv on sDM at least partially sharesthe DR-binding surface on DM strongly suggested by prior work(26, 37). Using mutant molecules for analysis, including FRETanalysis after dye addition, introduces uncertainty into both theDM/DR and the DM/DO docking models. We cannot rule outsubtle differences in the topology of these complexes; alteringthe region around DMβ141,142 confers some resistance to DOinhibition but has not been implicated in DR interaction.

0 500 1000 0 4000 8000 0 500 1000 0 500 1000 nM

sDMwt R110S H141A,S142A E177N,I179T

nm S

hift

(nm

)M

ax n

m S

hift

(nm

)

KD=1.74±0.44nM

R2=0.9692

KD=55.5±7.76nM

R2=0.9837

KD=10.1±2.34nM

R2=0.9793

KD=1.20±0.29nM

R2=0.9705

B

D top view side view

DM DO DO

sDM R110S

0 10000 20000Time (sec)

nm S

hift

(nm

)nm

Shi

ft (n

m)

1.0

0.5

0.0

1.0

0.5

0.0

0 10000 20000Time (sec)

sDM R110SF

old

chan

ge o

ver

400

nM

DM

, no

szD

Ov

[DO]

***

***

0

0.2

0.4

0.6

0.8

1

1.2

!"#$%& !"#'()*)+,&-)*.+& !"#'/)001,&2)034&

**

sDMwt sDM H141A,S142A sDM E177N,I179

A1.2

0.6

0.0Fo

ld c

hang

e ov

er40

0 nM

sD

M, n

o sz

DO

vsDMwt

H141A,S142A

E177N,I179T

[DO]

******

C

R80A62*

C46

R80A62*S63

C46K39E41

E131T128

H141E177

R110

K38

S142I179

S63

DM

K39E41

K38

sDMwt

Fig. 4. Topology of DO/DM complex. (A) Fast dissociation of sDMβR110Sfrom szDOv. Octet binding curves of sDMwt or sDMβR110S (both black) toimmobilized szDOv-biotin. Control curves (no immobilization, no sDM) weresubtracted. szDOv-biotin immobilization and no sDM is shown in red. (B)Octet analysis of sDMwt, sDMβR110S, sDMβH141A,S142A, and sDMβE177N,I179T binding to immobilized szDOv-biotin. (Upper) sDMwt, sDMβH141A,S142A, and sDMβE177N,I179T [at 0.24 nM (blue), 0.98 nM (brown), 3.9 nM(pink), 15.6 nM (cyan), 62.5 nM (sky blue), 250 nM (red), and 1,000 nM (black)]were used to determine KD, and concentrations up to 8,000 nM sDMβR110Swere used to reach saturation binding [15.6 nM (brown), 62.5 nM (pink),250 nM (cyan), 1,000 nM (sky blue), 4,000 nM (red), and 8000 nM (black)].(Lower) Plots of maximum wavelength (nm) shift under equilibrium bindingconditions against increasing concentrations of sDM proteins. (C) Effects ofDM mutations on szDOv inhibition of DM. szDOv was incubated withsDMwt, sDMβH141A,S142A, and sDMβE177N,I179T for peptide-loading as-say, as in Fig. 1A. Background loading in the absence of sDM is in left-mostbar with each set. Other bars represent data expressed as normalized foldchange, with enhancement of peptide loading in the presence of eachpreparation of sDM alone normalized to 1. Data, from one of threeexperiments with similar results, are shown. Statistics were obtained by two-way ANOVA, compared with sDMwt. **P < 0.01; ***P < 0.001. (D) Orien-tation of the szDOv/sDM complex. Top view (Left) and side view (Right) ofthe molecules: DMα and DOα (cyan); DMβ and DOβ (purple). Mutant residuesfrom all experiments are indicated, with the exception of DOαS81A, whichwas too aggregated for study. Color code is as follows: red for residues la-beled by dye addition for FRET (DO αA62, αR80, αE131, and βS63; *FRETanalysis places DOαA62 closest to DM); and brown for residues altered formutational analysis. Note: αT128N leads to glycan at αN128. βR110S leads toglycan at βN108. βE177N, I179T leads to glycan at βN177. Large font indicatesresidues providing data for proposed topology of the DO/DM complex; smallfont indicates residues that did not influence DO/DM interaction.

11280 | www.pnas.org/cgi/doi/10.1073/pnas.1113966109 Yoon et al.

Similarly, the carbohydrate adduct in mutant R110S may bemobile enough to influence contiguous, but nonidentical, sites.Nonetheless, given the strong sequence homology between DOand DR (38, 39), the most parsimonious interpretation would bethat a similar, if not identical, site on DM is used to bind DO andDR and that szDOv might directly interfere with the DR/DMinteraction to inhibit DM function.In EBV-transformed human B-cell lines expressing mutant

DMβ R110S (aberrant N-glycan at β108), levels of coimmuno-precipitated DO/DM complexes are reduced (26). Consistentwith this observation, sDMβR110S has reduced apparent affinityfor szDOv, arguing for similarity between the szDOv/sDM in-teraction and the interaction of full-length DO and DM mole-cules in cells. In contrast, two groups have reported a triplecomplex (DR/DM/DO) coimmunoprecipitated from cell lines,implying different DM-binding surfaces for DR and DO (20, 23).The discrepancy between these data and the possibility that DOand DR share a binding site may indicate that higher-ordercomplexes combining DR/DM and DO/DM are formed in cells,perhaps with contributions from other proteins. Alternatively,there may be differences between recombinant soluble proteinsand full-length proteins containing transmembrane and cyto-plasmic domains. These domains may mediate intermolecular

interaction between DM and DO, as shown for other membraneproteins (40). If so, these interactions influence the orientationsof the intraluminal domains of full-length molecules, which maydiffer to some degree from the model we present here and allowtriple complexes. The availability of recombinant forms of DOthat can be produced efficiently should accelerate the progress instructural and kinetic studies of DO and DO/DM complexes totest our model further.

Materials and MethodsRecombinant soluble DM expression and purification was performed asdescribed (41). Cloning, expression, and purification of recombinant solubleDO is described in SI Materials and Methods.

All other experimental methods are also described in SI Materialsand Methods.

ACKNOWLEDGMENTS. We thank Drs. J. Thibodeau, L. Karlsson, and L. Den-zin for reagents; Laura Su for sHLA-DR4 protein; and Guoqi Li for cys-labeledsHLA-A2. This work was supported, in part, by National Institutes of Health(NIH) Grants AI-095813 (to E.D.M.) and AI-38996 (to L.J.S.). H.M. and C.H.R.were supported by NIH Training Grant T32 AI07290 (to the Stanford Inter-disciplinary Program in Immunology). H.M. and W.J. were supported by theNational Center for Research Resources and the National Center for Advanc-ing Translational Sciences, National Institutes of Health, through UL1RR025744, and by the Lucile Packard Foundation for Children’s Health,and S.E.M. was supported by NIH Grants T32 AI07349 and F32 AI072984.

1. Busch R, et al. (2005) Achieving stability through editing and chaperoning: Regulationof MHC class II peptide binding and expression. Immunol Rev 207:242–260.

2. Bakke O, Dobberstein B (1990) MHC class II-associated invariant chain containsa sorting signal for endosomal compartments. Cell 63:707–716.

3. Lotteau V, et al. (1990) Intracellular transport of class II MHC molecules directed byinvariant chain. Nature 348:600–605.

4. Roche PA, Cresswell P (1991) Proteolysis of the class II-associated invariant chaingenerates a peptide binding site in intracellular HLA-DR molecules. Proc Natl Acad SciUSA 88:3150–3154.

5. Denzin LK, Cresswell P (1995) HLA-DM induces CLIP dissociation from MHC class IIalpha beta dimers and facilitates peptide loading. Cell 82:155–165.

6. Morris P, et al. (1994) An essential role for HLA-DM in antigen presentation by class IImajor histocompatibility molecules. Nature 368:551–554.

7. Sloan VS, et al. (1995) Mediation by HLA-DM of dissociation of peptides from HLA-DR.Nature 375:802–806.

8. Weber DA, Evavold BD, Jensen PE (1996) Enhanced dissociation of HLA-DR-boundpeptides in the presence of HLA-DM. Science 274:618–620.

9. Mosyak L, Zaller DM, Wiley DC (1998) The structure of HLA-DM, the peptide exchangecatalyst that loads antigen onto class II MHC molecules during antigen presentation.Immunity 9:377–383.

10. Kropshofer H, et al. (1996) Editing of the HLA-DR-peptide repertoire by HLA-DM.EMBO J 15:6144–6154.

11. Denzin LK, Hammond C, Cresswell P (1996) HLA-DM interactions with intermediatesin HLA-DR maturation and a role for HLA-DM in stabilizing empty HLA-DR molecules.J Exp Med 184:2153–2165.

12. Denzin LK, Sant’Angelo DB, Hammond C, Surman MJ, Cresswell P (1997) Negativeregulation by HLA-DO of MHC class II-restricted antigen processing. Science 278:106–109.

13. van Ham SM, et al. (1997) HLA-DO is a negative modulator of HLA-DM-mediated MHCclass II peptide loading. Curr Biol 7:950–957.

14. Liljedahl M, et al. (1998) Altered antigen presentation in mice lacking H2-O. Immunity8:233–243.

15. Alfonso C, Williams GS, Karlsson L (2003) H2-O influence on antigen presentation inH2-E-expressing mice. Eur J Immunol 33:2014–2021.

16. van Lith M, et al. (2001) Regulation of MHC class II antigen presentation by sorting ofrecycling HLA-DM/DO and class II within the multivesicular body. J Immunol 167:884–892.

17. Xiu F, et al. (2011) Cutting edge: HLA-DO impairs the incorporation of HLA-DM intoexosomes. J Immunol 187:1547–1551.

18. Yi W, et al. (2010) Targeted regulation of self-peptide presentation preventstype I diabetes in mice without disrupting general immunocompetence. J Clin Invest120:1324–1336.

19. Draghi NA, Denzin LK (2010) H2-O, a MHC class II-like protein, sets a threshold forB-cell entry into germinal centers. Proc Natl Acad Sci USA 107:16607–16612.

20. Kropshofer H, et al. (1998) A role for HLA-DO as a co-chaperone of HLA-DM inpeptide loading of MHC class II molecules. EMBO J 17:2971–2981.

21. van Ham M, et al. (2000) Modulation of the major histocompatibility complex classII-associated peptide repertoire by human histocompatibility leukocyte antigen(HLA)-DO. J Exp Med 191:1127–1136.

22. Liljedahl M, et al. (1996) HLA-DO is a lysosomal resident which requires associationwith HLA-DM for efficient intracellular transport. EMBO J 15:4817–4824.

23. Deshaies F, et al. (2005) A point mutation in the groove of HLA-DO allows egress fromthe endoplasmic reticulum independent of HLA-DM. Proc Natl Acad Sci USA 102:6443–6448.

24. Deshaies F, et al. (2009) Evidence for a human leucocyte antigen-DM-induced struc-tural change in human leucocyte antigen-DObeta. Immunology 127:408–417.

25. Stern LJ, et al. (1994) Crystal structure of the human class II MHC protein HLA-DR1complexed with an influenza virus peptide. Nature 368:215–221.

26. Pashine A, et al. (2003) Interaction of HLA-DR with an acidic face of HLA-DM disruptssequence-dependent interactions with peptides. Immunity 19:183–192.

27. Mujumdar RB, Ernst LA, Mujumdar SR, Lewis CJ, Waggoner AS (1993) Cyanine dyelabeling reagents: Sulfoindocyanine succinimidyl esters. Bioconjug Chem 4:105–111.

28. Gruber HJ, et al. (2000) Anomalous fluorescence enhancement of Cy3 and cy3.5 versusanomalous fluorescence loss of Cy5 and Cy7 upon covalent linking to IgG and non-covalent binding to avidin. Bioconjug Chem 11:696–704.

29. Muñoz-Losa A, Curutchet C, Krueger BP, Hartsell LR, Mennucci B (2009) Frettingabout FRET: Failure of the ideal dipole approximation. Biophys J 96:4779–4788.

30. Denzin LK, Fallas JL, Prendes M, Yi W (2005) Right place, right time, right peptide: DOkeeps DM focused. Immunol Rev 207:279–292.

31. Määttänen P, Gehring K, Bergeron JJ, Thomas DY (2010) Protein quality control in theER: The recognition of misfolded proteins. Semin Cell Dev Biol 21:500–511.

32. Van den Steen PE, Opdenakker G, Wormald MR, Dwek RA, Rudd PM (2001) Matrixremodelling enzymes, the protease cascade and glycosylation. Biochim Biophys Acta1528:61–73.

33. Dustin ML, et al. (1997) Low affinity interaction of human or rat T cell adhesionmolecule CD2 with its ligand aligns adhering membranes to achieve high physio-logical affinity. J Biol Chem 272:30889–30898.

34. Ryan SO, Bonomo JA, Zhao F, Cobb BA (2011) MHCII glycosylation modulates Bac-teroides fragilis carbohydrate antigen presentation. J Exp Med 208:1041–1053.

35. van Lith M, Benham AM (2006) The DMalpha and DMbeta chain cooperate in theoxidation and folding of HLA-DM. J Immunol 177:5430–5439.

36. Doebele RC, Busch R, Scott HM, Pashine A, Mellins ED (2000) Determination of theHLA-DM interaction site on HLA-DR molecules. Immunity 13:517–527.

37. Stratikos E, Mosyak L, Zaller DM, Wiley DC (2002) Identification of the lateral in-teraction surfaces of human histocompatibility leukocyte antigen (HLA)-DM withHLA-DR1 by formation of tethered complexes that present enhanced HLA-DM ca-talysis. J Exp Med 196:173–183.

38. Servenius B, Rask L, Peterson PA (1987) Class II genes of the human major histo-compatibility complex. The DO beta gene is a divergent member of the class II betagene family. J Biol Chem 262:8759–8766.

39. Kelly AP, Monaco JJ, Cho SG, Trowsdale J (1991) A new human HLA class II-relatedlocus, DM. Nature 353:571–573.

40. Matsumoto AK, et al. (1993) Functional dissection of the CD21/CD19/TAPA-1/Leu-13complex of B lymphocytes. J Exp Med 178:1407–1417.

41. Busch R, Reich Z, Zaller DM, Sloan V, Mellins ED (1998) Secondary structure compo-sition and pH-dependent conformational changes of soluble recombinant HLA-DM.J Biol Chem 273:27557–27564.

Yoon et al. PNAS | July 10, 2012 | vol. 109 | no. 28 | 11281

IMMUNOLO

GY